JP2009104602A - Real-time query suggestion method and system in troubleshooting context - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To allow a customer to judge how to repair an electromechanical device without requiring a visit from a service engineer when the customer observes a problem with the device. <P>SOLUTION: The method includes storing logs comprising information derived from prior user search sessions in which user queries are input to a search engine for retrieving responsive instances from a knowledge base. A collection of query suggestions is stored. Each of the query suggestions is formulated to retrieve at least one responsive instance in the knowledge base. The query suggestions in the collection are ranked, based at least in part on the stored logs. A user's query is received in a natural language. While the user's query is being entered, a subset of the ranked collection of query suggestions is generated and one or more of the query suggestions in the subset are presented to the user as candidates for user queries. The selection of the subset is based on that portion of the user's query already entered. The presentation of the query suggestions in the subset of query suggestions is based on their respective rankings in the collection. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  Exemplary embodiments according to the present invention relate to language processing techniques. It helps users to develop queries for searching structured knowledge bases such as manuals to troubleshoot electromechanical device failures. A specific application associated with is found and will be described with specific reference. However, it can be modified for other similar applications.

  Many of the devices used today both inside and outside the workplace are highly complex. Such devices are electromechanical devices such as image reproduction devices such as printers and copiers, vehicles, audiovisual equipment (such as cameras, video recorders, mobile phones, dictation devices, music systems), computing devices such as personal computers, etc. including.

US Patent Application Publication No. 2007/0192085 US Pat. No. 6,564,213 US Pat. No. 7,058,567 US Pat. No. 4,839,853

  When a customer observes a problem in an electromechanical device, it is advantageous for the customer to determine how to repair the device without requiring a visit from a repair technician. However, problem solving systems using query-based user interaction in natural language typically have some expertise in terms of terminology used in the knowledge base on which the user is based. Provide efficient results only if “good” words can be selected in (query).

  Furthermore, a device without a suitable keyboard makes query entry and refinement difficult.

  According to one aspect of an exemplary embodiment, a method is provided for assisting a user in developing a query in natural language. The method includes storing a log that includes information derived from a previous user search session in which a user query (query) was entered into a search engine to retrieve a response instance from the knowledge base. A set of inquiry proposals (suggestions) is stored. Each query proposal is formulated to search for at least one response instance in the knowledge base. Query suggestions in the set are ranked based at least in part on the stored logs. A user query is received in the form of a natural language. While a user query is being entered, a subset of the set of ranked query suggestions is generated, and one or more of the query suggestions in the subset is selected. The above is presented to the user as candidates for the user inquiry (query). The selection of the subset is based on the already entered part of the user's query. The presentation of query proposals in a subset of query proposals (suggestions) is based on the respective ranking of the query proposals (suggestions) in the set.

In various embodiments,
Each response instance contains a string.
Presenting a query suggestion includes presenting a linguistically consistent representation with respect to at least one of the query suggestions.
The linguistically consistent representation is at least one of the representations that are the same as, synonymous with, or have the same normalized form as the representation of at least one response instance in the knowledge base.
The user query may be related to troubleshooting a problem in the device, and each instance of the knowledge base includes at least a known problem in the device and a known solution to the known problem. A character string representing one may be included.
Each query suggestion can have a surface form, which is a linguistically consistent representation suitable for presentation to the user, and presents a query suggestion Includes presenting at least one surface form of a query proposal in the subset.
At least some surface forms of the query suggestions are derived from previous user queries stored in the log.
Ranking is based at least in part on the frequency with which query suggestions are found in stored logs.
Ranking includes determining for each of the query suggestions (suggestions) how often a normalized form of meaningful words in the query (suggestions) appears in the log. And ranking is based on the determined frequency.
Storing the log can include, for each of a plurality of user queries, storing the context in which the user query (query) occurred in a respective session, the context being a plurality of predetermined Selected from the context.
The ranking of a set of query suggestions (suggestions) may be based at least in part on the stored context.
The ranking of the query proposals (suggestions) may be based at least in part on the relationship between one of the query proposals (suggestions) and the knowledge base.
The ranking of each query proposal (suggestion) may be based at least in part on the score of the response instance in the knowledge base.
Presenting a subset of a set of ranked query suggestions to a user as a candidate for a user query (query) can result in subsets as more user queries (queries) are entered. Including updating.
The method further includes allowing a user to accept a query suggestion as a user query and querying the knowledge base based on the user query.
The method further extracts a query proposal (suggestion) candidate from the knowledge base, and a surface form suitable for presenting the user with the query proposal (suggestion) candidate having a uniform normalized form. Grouping into a single query suggestion.
If one of the query proposals in the set is a synonym for another query proposal in the subset, the method can use a query proposal that is a synonym ( Suggesting (suggestions) with a query suggestion (suggestions) in which it is synonymous.
Presenting includes displaying on the graphical user interface the surface form of at least the highest ranked subset of the subset of query suggestions.
A computer program product has encoding instructions that, when executed on a computer, cause the computer to perform the method.

  According to another aspect of the exemplary embodiment, a system for assisting a user in developing a query in natural language (search) each retrieves at least one response instance in the knowledge base. A memory for storing a set of query proposals formed. The ranking of query suggestions in the set is based at least in part on stored logs of previous user interactions. The troubleshooting session support module in communication with memory receives the user's query in natural language and ranks the query while the user's query is entered (query). Query) Present at least one of a subset of a set of suggestions to a user as a candidate for a user query. The subset is based on the already entered part of the user's query. Query (suggestion) suggestions in a subset of query (suggestion) suggestions are presented by respective rankings of query (query) suggestions (suggestions) in the set.

In various embodiments,
The system further includes a module for collecting and ranking query suggestions that ranks a set of query suggestions based at least in part on the stored logs.
A module that collects and ranks query proposals (suggestions) extracts candidate queries (suggestions) from the knowledge base.
At least some of the query proposals in the set may each be just a synonym for an expression in one of the response instances in the knowledge base, which includes the expression Is given the same ranking as another query proposal that has the same normalization form.

  The ranking of query suggestions in the set is stored in a stored log of previous user sessions where user queries (queries) were entered into the search engine to find response instances from the knowledge base. Based at least in part.

  Various aspects of the exemplary embodiments allow linking between the user's terminology and knowledge-based terminology and also provide immediate feedback for use in problem solving and other contexts To.

  In an exemplary embodiment, a set of ranked, pre-determined query suggestions (suggestions) (suggestions for user queries (queries)) derived from a knowledge base is stored in memory. . When the user types a new character in the query input box of the user interface, the corresponding query suggestion is retrieved and presented according to the ranking in the set. Ranking gives an estimate as to the likelihood that each query will successfully retrieve results corresponding to the user's intended query. An estimate of the likelihood that the query proposal used to generate the ranking will be successful is signaled by two considerations. The first consideration is the quality of the results that match the query in the knowledge base. This helps to suggest queries with high quality results. The second consideration is based on the history of previous problem solving sessions. This helps suggest a query that may match what the user wants to type.

  Since a set of queries that are candidates for suggestions (suggestions) is created from the searched knowledge base, only query suggestions (suggestions) that retrieve at least one result from the knowledge base are considered. Be put in. The query suggestions in this set are then ranked according to their history (logs from previous troubleshooting sessions). This ranking is at least in part a) frequency, eg how frequently they were used in previous problem solving sessions, b) performance, ie those in previous problem solving sessions. Based on how well it worked. Furthermore, other factors such as the frequency of occurrence in the knowledge base can be used as a factor for ranking query proposals (suggestions).

  Each query suggestion can include a linguistically consistent expression that includes a single word or group of words that are syntactically related. Examples of such expressions include noun phrases, prepositional phrases, verb syntax units, and the like. Each query suggestion may have a surface form of linguistically consistent representation suitable for presentation to the user and an underlying (normalized) form. At least one instance of each query proposal (suggestion) exists in the knowledge base in the form of a surface form, a normalized form, or a specified synonym of the query (query) proposal (suggestion); Thus, a query suggestion can be searched for instances when entered into a search engine.

  FIG. 1 illustrates a functional block diagram of an exemplary system 10 for developing a query suggestion. The system 10 executes instructions for retrieving a query suggestion corresponding to a query that the user has at least partially developed. The system 10 can be implemented in a personal / company computing system, or a device such as a printer, and / or can be implemented elsewhere such as a server. The illustrated system 10 includes a knowledge base 12, a set of logs 14 (actual search data or data developed (developd) therefrom) and a log 14 developed (development) when the knowledge base was previously searched. And a set 16 of query proposals (suggestions) developed from the knowledge base in conjunction with the ranked ranking. The system 10 also generates a set 16 of ranked query proposals (suggestions), and optionally updates it periodically based on information from the log 14 and knowledge base 12. Query) Includes a module 18 that collects and ranks suggestions. The problem solving session support module 20 develops a set of query proposals from the set 16 in response to a query entered by the user. In the illustrated embodiment, the support module 20 includes a type input component 22 that updates the response during the typing of a query by the user. The type input component 22 filters the set of ranked query proposals (suggestions), in particular, to identify a subset of query suggestions that the user has already typed. The type input component presents one or more of the query (suggestion) suggestions in the subset to the user based on the ranking of the query (suggestion) suggestions in the subset.

  FIG. 2 illustrates, by way of example, an environment in which an exemplary system can operate. The knowledge base 12, log 14, and ranked query proposal set 16 may be stored in a memory 24. The system 10 may further include a regional spell check component 26 that is accessible within or accessible to the support module 20. All components 12, 14, 16, 20, 22, 24, 26 of the system 10 can be interconnected by one or more data / control buses 28.

  The processing components 18, 20, and 22 of the system 10 can be embodied in software, hardware, or both. In one embodiment, components 20 and 22 execute instructions stored in memory 24 to perform the exemplary methods described herein and illustrated in FIGS. 3 and 4. Thus, the operation of these components is only briefly described here.

  The system 10 interacts with a user interface 30 that resides on a device 32, illustrated in the illustrated embodiment as a multi-function device (MFD). The system 10, or a portion thereof, can reside on the server 34 or at another location remote from the device 32 and can be a computer network, such as a local area network (LAN) or a wide area network (WAN). The user interface 30 can be communicated via a wired or wireless link 36, such as the Internet. In another embodiment, system 10, or components thereof, may reside on device 32.

  The device 32 is a network computing device such as a printer, copier, scanner, fax machine, multifunction device, network router, network switch, server or workstation, network file system, heating or cooling system, refrigerator. It can be a household appliance such as a washing machine or dryer, or any other electromechanical device having an electrical or mechanical part that requires repair, replacement or modification. The device may be connected to the communication network 36.

  The exemplary device 32 may include a number of actuating components (not shown) that may sometimes contribute to operational failure of the device individually or in combination. Such a failure may be observed as a problem by a worker using the apparatus. For example, in the case of a printer, the part typically has one or more paper feeders and a colorant, such as ink or toner, applied to a paper or other print media substrate to form an image thereon. Optional image application parts or marking engines, and optionally a fusing system for more permanently fixing the image to the substrate, a finishing machine, and a conveyor system for sequentially transporting the print media from the paper feeder to other parts including. The device may further include whether or not at least some troubleshooting attempts by the user (eg opening or closing certain doors, removing certain parts) and / or certain detected faults have been corrected. Can include or communicate with one or more internal diagnostic systems 38 that can detect whether it has been corrected, such as by using a sensor. it can.

  The system 10 further includes or includes a search engine 40 that forms a user query in a format suitable for querying the knowledge base to retrieve response instances. Can communicate. Optionally, the thesaurus 42 stores synonyms of words or expressions that have been recognized in the knowledge base.

  The user interface 30 may be a graphical user interface (GUI), a display 50 such as a screen, and a user input device 52 such as a touch screen, keyboard, keypad, cursor control device, or combinations thereof; Can be included. The user interface can interact with the problem solving system 10 via a web browser displayed on the display 50.

  As shown in FIG. 5, the display 50 includes a user query (query) box 54 that displays an input user query, and a ranked query (development) developed by the support module 20 (development). Query) a suggestion area 58 for automatically displaying one or more of the suggestions. Although not shown, the display 50 may also display a response to the user's query, e.g., expand the query (query) to more closely match the problem description in the knowledge base (e.g., via a navigation tree). Options for developing may be displayed. The user interacts with the user interface 30 by manipulating the associated user input device 52. A user may enter a query as a string, and further features typically associated with screens and toolbars, popup windows, scroll bars, menu bars, drop-down menus, and other web browsers. Other graphical user interface features, such as one or more of the above, may be navigated. The user interface 30 inputs a complete or partial query to the system 10 via a suitable input / output device 54.

  The user interface 30 is a dedicated computer (eg, a digital front end of a printer), a programmed microprocessor or microcontroller such as a computer network workstation, a mobile computer device such as a PDA, a mobile phone, a tablet PC Or using a pocket PC. Memory 24 may represent any type or combination of computer readable media such as random access memory (RAM), read only memory (ROM), and the like.

  Knowledge base 12 can include any closed corpus (typically a limited set of one or more digital documents written in a natural language such as English or French) in the form of a relational database. Or as a hypertext document such as an Extensible Markup Language (XML) document that is indexed, for example, annotated with tags or other structured formats that index content Also good. The illustrated knowledge base 12 that can be stored in the memory 24 on the server includes a set of annotated strings that may be retrieved as corresponding to a user query. Specifically, the strings in the knowledge base 12 can take the form of defined cases, each describing a possible problem with the device 32. Each problem is associated with a set of solutions in the knowledge base that may fix the problem. The knowledge base can be indexed by the normalized form of words that occur in the problem description, optionally by the solution or part thereof, and optionally by its synonyms, and by the syntactic unit that the description contains can do. The knowledge base can include any limited set of searchable text.

  Each syntactic unit can be a linguistically consistent expression that is a series of grammatically related words in the same sentence, such as a noun phrase, a prepositional phrase, a verb syntactic unit, a combination thereof, and the like. For example, from the problem description sentence “white line when copying from document glass and document feeder”, the expressions “white line”, “from document glass”, and “from document feeder” are extracted. , And can be used for inquiry proposals (suggestions).

  A module 18 that collects and ranks query proposals (suggestions) collects linguistically consistent expressions from the knowledge base and also includes previous user solutions (troubleshooting), including user queries (queries). A representation ranking is calculated based at least in part on the session log 14. Such a log 14 may include a log 14 of multiple user problem solving sessions for devices of the same classification. During a subsequent troubleshooting session, the type input module uses the dynamic query (suggestion) suggestion mechanism to enhance the typed input, suggesting queries. A set of (queries) can be used.

  In a typical problem solving session that is incorporated into the log 14, the user can type in a query about the device and send it to the search engine 40, which in turn has a knowledge base. A response instance such as a character string is searched from 12. The user can check the response instance and, if no instance is likely to be useful or if no instance exists, enter a new query or modify an existing query be able to. If the user chooses to read a document returned as containing a response instance, a record of which document was retrieved can be stored in the log of a problem solving session. Several user queries may be sent in a single session and may be identified in the log as coming from the same session. The log 14 can include sessions from multiple users on the same device or different devices (usually one of the same general type or classification).

  FIG. 3 shows a list of query suggestions for users who create a set of dynamic query suggestions, rank them, and type the queries. And how this list is dynamically updated during typing. The method starts at S100.

  In S102, query proposal suggestions are collected from the knowledge base.

  In S104, the query (suggestion) suggestion candidates collected from the knowledge base and the other query (suggestion) suggestion candidates that are synonyms thereof are stored in the set.

  In S <b> 106, a set of inquiry (query) proposals (suggestions) is generated from the inquiry (query) proposal (suggestion) candidates. Some of the query proposals (suggestions) may group several query (suggestion) suggestion candidates that have the same headword (normalized) form, most frequently occurring in the log 14 It may be displayed in a surface form of a query proposal (suggestion) (two or more surface forms are found for the same headword form of an expression). Expressions that are synonyms may each be linked to a query suggestion that is determined to be synonymous.

  In S108, the previous user interaction log is stored as a set 14 of logs. For example, the component that collects and ranks suggestions for the system 10 retrieves a pre-acquired log stored at an appropriate location in the memory 24 during a previous problem-solving session, and from the previous log , To extract relevant information for incorporation into the log set 14. The set of logs can include a list of entered, typed user queries (and / or word headword forms in each query), for each query (query), Whether the user reshaped (reformatted) the query in the session, whether the user entered one (query) and then examined one of the search results before reshaping the query (query) Information indicating the context of the query, such as whether or not, or if the diagnostic system 38 is applicable, for example, whether the system subsequently indicates that the fault has been corrected, or that the problem has been corrected Other information useful in the ranking implementation, such as whether the user reported or not, can be included.

  In S110, query proposals are ranked. Ranking a set of queries (suggestions) may be performed by a component 18 that collects and ranks suggestions (suggestions). In the exemplary embodiment, the ranking is based at least in part on information derived from stored logs 14 of previous problem solving sessions. In one aspect, the ranking is based on one or more criteria (frequency in logs and / or queries) based on information related to query suggestions derived from the log 14. ) And one or more criteria based on information related to query suggestions derived from the knowledge base 12 may be taken into account.

  In S112, the user initiates a user query (query), for example, by typing a query on the user interface. The system 10 is updated with queries typed by the user so that, for example, as each new character is typed during typing, it is received by the system. Thus, the system 10 does not wait for the user to respond until it sends a complete query.

  At S114, one or more query suggestions from the set 16 are presented to the user as candidates for the user query (query) while the user is typing in the type. The presented query proposal is based on the content of the user query that has already been entered. This includes filtering the set of query proposals (suggestions) by a set of rules to identify a subset of the corresponding query proposals (suggestions). The order in which the corresponding query suggestions are presented is based on the ranking of the query suggestions in the set 16.

  In S116, the user can accept one of the presented query proposals as a user query.

  In S <b> 118, a user query can be input to the search engine 40. The search engine can retrieve one or more instances of user queries (queries) (adopted queries (suggestions) suggestions) from the knowledge base and present them for selection by the user. When the user selects one of the instances, additional knowledge base content can be presented, such as at least one suggested solution stored in the knowledge base 12.

The system may present options for improving the user's query (query) by presenting the extended word before entering the query (query) selected by the user into the search engine 40. Each extended word is found in the same string as the user's query (query) in the knowledge base (which can be a query suggestion), a surface form of another linguistically consistent representation It can be. For example, if the user query is “white line”, the system
White lines when printing White lines when printing and faxing White lines when copying from document glass can be presented as an option for improvement.

  Steps S102 to S110 generally precede a problem solving (troubleshooting) session.

  In S120, a log of user interaction with the system, such as a problem solving (troubleshooting) session, is stored in log 14, and the ranking of query proposals (suggestions) in subsequent sets 14 is updated. Can be used. If the knowledge base 12 is subsequently updated, the set of query proposals (suggestions) can also be updated to reflect that.

  The method ends at S122.

  FIG. 4 illustrates exemplary sub-steps that can be performed in the step of presenting a query suggestion (S114) (S114). Specifically, for each character newly typed by the user in S200, a subset of the query proposal (suggestion) is generated in S202 based on the user's query typed so far. Is done. In this first stage, the entire user query typed so far is considered and only query suggestions that match the entire query are returned. Synonyms query suggestions can also be returned at this stage. In S204, if the subset is empty, the method continues to step S206, where a query proposal that matches only a portion of the query is returned. If the subset is empty in S208, the spelling of the query can be checked in S210 and the preceding steps S202, S204, and optionally S206 are repeated as necessary. If the subset is still empty at S212, no query is returned and the method returns to S200 and waits for the user to modify the query. If not empty, in S214, one or more of the highest ranked query suggestions are suggested as candidates for auto-complete the user query, and their synonyms in the vicinity. Can be placed and presented to the user. In the exemplary embodiment, the sub-step of S114 can be interrupted at any time by the addition of another letter by the user or modification to other user queries, in which case the method directly Return to S200.

  Further details of an exemplary method will now be given.

Proposal (Suggestion) candidate set (S102)
A set of candidates for a query suggestion can be constructed from a knowledge-based index. Specifically, the following elements of the index are relevant as a query proposal.
1. Syntax unit. Since the queries created from these are guaranteed to have an exact match, they are used as the primary source for query suggestions.
2. synonym. Synonyms are generally domain specific in that they are expressions that can be predicted to be used by a user in a search in a specific domain of the knowledge base.

であり、ここでＮはログに格納された照会（クエリ）提案（サジェスチョン）インスタンスの合計数であり、ｋ₁及びｋ₂は、一方のスコア又は他方のスコアを有利にするように定めることができる２つの重み付け値である。 Ranking inquiry (suggestion) (S106)
A number of criteria can be used to rank query suggestions, which may be used alone or in combination. Each query suggestion may be given a score S _c based on the frequency of instances in the knowledge base. A query suggestion can additionally or alternatively give a _Sl based on the occurrence in the log. These scores can be combined with an appropriate weighting to generate an overall score S _total for the query proposal. That is, the following equation 1:

Where N is the total number of query suggestion instances stored in the log, and k ₁ and k ₂ are defined to favor one score or the other score. There are two possible weight values.

Score based on log S _l
The log-based S _l may be based on the frequency of occurrence of query suggestions in the log (e.g. represented by the frequency in the headword form) and / or the performance of the query (query). It may be based on a measure (an estimate of the probability that a query suggestion will be successful).

To calculate the frequency f _l of query suggestions in the log, a meaningful word in the expression (ie avoiding very common words) is first identified and its headword (eg, “Print” is identified as the headword of “print”. This set of headwords is compared to the set of headwords calculated for each old user query stored in the log 14. If the set of meaningful headwords in the query (suggestion) expression is a subset of the set of headwords in the old user query (query), the query (suggestion) expression will appear in the log The frequency of query suggestions is considered to be incremented.

  If various instances of a query suggestion are found in the log, the query (query) in identifying and displaying the surface form of one query (suggestion) suggestion to the user It may be used as a representative of a proposal (suggestion) (S114).

  To evaluate performance, the location of the query instance in the session (instance context) can give an idea of how the query was successful. For example, if a query instance was not reformed (ie it was the last query (query) instance in a session), it is more than a reshaped query (query) instance. It can be regarded as likely to succeed. If other user interactions are stored in the log along with the query instance, they can also be used to estimate the probability that the query instance will succeed. For example, if there is (or does not) a user selects a solution page next to a query instance, this may indicate a higher (or lower) probability that the query (query) will succeed. In one embodiment, the probability of success of a query instance in the log can be used to weight the increment given to the frequency of suggestions that match the query instance.

  If information about failures and successes in past troubleshooting sessions is available, this information can be used to select only the latest queries from successful sessions, or at least suggestions ( They may be advantageous when calculating the frequency of suggestions.

によって与えることができ、ここで照会（クエリ）提案（サジェスチョン）Ｓは、語の組としてモデル化され、Ｑはログにおける照会（クエリ）インスタンスの組であり、各々の照会（クエリ）インスタンスＱ_iは、語の組としてモデル化される。用語がログにおける照会（クエリ）提案（サジェスチョン）のすべての語を含む照会（クエリ）インスタンスの各々に関して適用される関数Ｗとして定められる場合に、照会（クエリ）提案（サジェスチョン）のスコアＳ_lｓは合計として算出される。Ｓ_lｓは、上記の数（１）におけるＳ_lとして用いてもよく、又は他のログに基づく基準に関する他のスコアと組み合わせてもよい。 In one embodiment, a query proposal is proposed by summing the weights based on the probability that a particular instance will produce a successful result across all instances of the query proposal in the log 14. A plurality of criteria can be integrated in the calculation of the score S ₁ s. Such a calculation is

Where the query proposal S is modeled as a set of words, Q is the set of query instances in the log, and each query instance Q _i Is modeled as a set of words. When the term is defined as a function W applied for each query (query) instance containing all the words of the query in the log (query suggestions), the score S _l s of query suggestions Is calculated as a sum. S _l s may be used as S _l in number (1) above, or may be combined with other scores for other log-based criteria.

である。 Function W models the probability of success of a query instance given a context in the log. One way to define this function is as follows:

It is.

Score S _c based on knowledge base
The score S _c based on the content of the knowledge base may also be a function of one or more criteria such as a measure of relative importance r _{c of} queries (suggestions) in the knowledge base content. it can.

In one embodiment, query proposals (suggestions) have a higher r _c , for example (assuming a higher weight is given to the title in the ranking system) they are in the title of the document. If it occurs, you are given a higher score.

The score S _c may additionally or alternatively be a function of the frequency f _c at which a query suggestion occurs in the knowledge base, or the query suggestion (suggestion) It may be a function of the number of problem descriptions retrieved by querying the knowledge base.

Dynamic update of inquiry (query) suggestion list (S114)
The sequence of characters already typed in by the user can be used to filter a set of ranked query suggestions. The best ranked suggestions in the subset of query suggestions that result from filtering are displayed in an area 58, for example, below the type entry area 54, so that the user can always One of the query proposals (suggestions) can be adopted to shorten typing.

In filtering performed on a set of query (query) suggestions, the corresponding query (suggestion) suggestions may meet at least one of the following criteria: That is,
a) Includes all typed words and any partial words.
b) The normalized form includes all of the typed words and any normalized form of any subword.
c) Designated as a synonym for another query (suggestion) that is determined to correspond.
d) The normalized form includes at least one and a minority of the typed word and any subword normalized form.
e) Satisfy one of the above criteria a), b), c), d) in a corrected form (eg by using an area-specific spell checker).

  In one embodiment, one or both of criteria d) and criteria e) is such that the previous criteria does not return a query proposal or a limited number of query suggestions. It can be activated only when it returns only (S206 and S210).

  To perform the filtering, the user query (query) is first tokenized to a word or, for the rightmost part, a word prefix. These words (or the entire word) are then made headlines. Finally, a search is performed over a set of query suggestions. Contains a query (query) word (or prefix) set (e.g., criteria a), b), c) meet one or more of the criteria, and optionally criteria c) if no other criteria succeed Or all queries (suggestions) that satisfy e) are returned as search results.

  For example, in the context of an MFD problem solving (troubleshooting) scenario, if the user types “li”, the filtered query proposal that is ranked highest is “line (line)”. ) "," Light "," list ", etc., but may include single word expressions that begin with" li ", but one word (not necessarily the first word) is" lighter " In some cases, a multi-word expression starting with “li”, such as “copyes (brighter copy)” or “black lines (black line)”. In another example, when the user has typed an entire word and typed a second word, such as “feeder t”, the corresponding query suggestion is , “Document feeder tray”, “high capacity feeder tray”, “document feeder top cover”, etc. .

  If the user input contains several words and a suggestion for the entire representation is not available, one option is to propose a suggestion for the longest and / or latest part of the input item (Suggestion) (S206).

In an exemplary embodiment, synonym suggestions are processed in a manner that is somewhat different from other query suggestions. Such a query suggestion is displayed both when it matches a set of characters already typed by the user and when it is a synonym for the matching expression. For example, if the user types “j”, the query proposal in the MFD context is
jam paper jam multifeed multifeed (multiple feed)
misfeed misfeed job It looks like a job and “misfeed” and “multifeed” are stored in the thesaurus as synonyms for “paper jam”, so they appear just below jam (paper jam) . If there is no query proposal for user input, you can use a spell checker to check for typing mistakes, and query suggestions as a way to correct it. (Suggestion) can be used (S210). Suggested corrections are taken from a limited word list that is part of the knowledge-based corpus.

  The methods illustrated in FIGS. 3 and 4 can be implemented in a computer program product that can be executed on a computer. The computer program product can be a tangible computer readable recording medium, such as a disk, hard drive, or any other medium from which a computer can read and use, on which a control program is recorded.

  In general, any device that can implement a finite state machine, i.e., can then implement the flowcharts of FIGS. 3 and 4, forms a query for a user to enter into a search engine. It can be used to implement a method that assists in (formulation).

  5, 6 and 7 show a GUI 30 for entering a search query that is displayed to the user. In the exemplary embodiment, the MFD device hosting the GUI does not have a complete alphanumeric keyboard, so a soft keyboard for typing queries is available to the user. The left side of the screen is used to display a query typed by the user, and below it is a suggestion for completion provided by the system.

  FIG. 6 illustrates the effect of “pressing” the letter l on a soft keyboard. The letter l is added to the query box, and a query query suggestion containing one or more words starting with “l” is displayed below. The user can select one of the query suggestions in the list at any point in time, and the contents of the query suggestion box will contain the selected query suggestion. It is updated using (suggestion). If the user continues to type, the query suggestion is shown in FIG. 7, which shows the user the highest ranked query suggestion for the user input “line”. As such, it is adapted to queries.

The exemplary system can combine the following features:
A query proposal is a representation constructed from a linguistic analysis of the corpus.
A query suggestion is suggested to the user as each character of the query is typed. Query proposals (suggestions) are ranked taking into account their use in prior sessions and, optionally, the context of occurrence in the knowledge base.

  The exemplary embodiment does not always suggest additional words for the system to add to the query (suggestions), but instead of an alternative query (query) that need not necessarily be an extension of the original word. ) In contrast to query expansion in that it suggests a system and method for query suggestions. This is commonly used in search engines with AND semantics, i.e., to be considered a match, the document (or instance) must contain all of the terms of the query, and thus the query (query). Adding words to) reduces the number of results.

  It will be appreciated that the various features and functions disclosed above and others, or alternatives thereof, can be desirably combined into many other different systems or applications. Further, those skilled in the art will later make various alternatives, modifications, variations or improvements which are not currently foreseen or anticipated which are also intended to fall within the scope of the appended claims. I can do it.

1 is a functional block diagram of a system for developing a query suggestion according to a first aspect of an exemplary embodiment. FIG. FIG. 2 is a diagram of an environment in which the exemplary system of FIG. 1 operates. 6 is a flow diagram illustrating a method for developing a query suggestion according to a second aspect of an exemplary embodiment. FIG. 4 is a diagram of substeps of the exemplary method of FIG. FIG. 3 is a representation of an exemplary query type input interface that can be generated by the system of FIGS. 1 and 2. FIG. 4 is a diagram of starting a user query (query) in the query type input interface of FIG. 3. It is a figure of the inquiry (query) proposal (suggestion) after the further type input was performed.

Explanation of symbols

10 System 12 Knowledge Base 14 Log 18 Query Proposal Collection and Ranking Component 16 Ranked Set 20 Problem Solving Session Support Component 22 Type Input Component

Claims (3)

A method for assisting a user in developing (developing) a query in natural language,
Stores a log containing information derived from previous user search sessions in which user queries (queries) were entered into the search engine to search for response instances from the knowledge base;
Storing a set of query suggestions each formed to search for at least one response instance in the knowledge base;
Ranking the query proposals in the set based at least in part on the stored logs;
Receive user queries in natural language,
Generating a subset of the set of ranked query suggestions while the user's query is being entered, and presenting at least one of the subsets to the user as a candidate for a user query. ,
The subset is based on an already entered portion of the user's query;
The presentation of query proposals in the subset of query proposals is based on a ranking of each of the query proposals in the set;
A method characterized by that.   The method of claim 1, wherein the ranking is based at least in part on the frequency with which query suggestions are found in the stored log. A system for assisting a user in developing (developing) a query in natural language,
A memory for storing a ranked set of query proposals (query suggestions), each of the query proposals being configured to retrieve at least one response instance in a knowledge base; Ranking the query proposals is based at least in part on a stored log of previous user interactions;
And a problem solving session support module in communication with the memory, wherein the problem solving session support module receives a user query in natural language, and the user query is input. During which at least one of the subset of the set of ranked query proposals is presented to the user as a candidate for a user query;
The subset is based on an already entered portion of the user's query, and in the subset of query proposals, the query proposal is presented by a respective ranking of the query proposals in the set;
A system characterized by that.

Images